RiddhiCh2027MIT
SGE-Goal CALVIN-ABCD
Training and evaluation dataset for spatially-grounded goal-image synthesis for robotic manipulation, extracted from CALVIN environments. Each sample converts an atomic manipulation step into an image-editing example with before/after scenes, edit masks, and instructions.
Downloads51
Episodes24053
Why This Matters for Physical AI
This dataset bridges vision-language models and robotic manipulation by providing automatically-extracted goal-image supervision for training spatially-grounded manipulation policies from language instructions.
Technical Profile
- Modalities
- rgblanguage
- Action Space
- language
- Environment
- simulation
- Task Types
- manipulationimage-to-imagegoal-image-generation
- Episodes
- 24053
- Annotation Types
- language_instructionssegmentationbounding_boxes
- License
- MIT
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack